We are collecting our Data from the sustainability development report (SDG), the international labour organization (ILOSTAT), the World Bank, Our world in data, the CATO institute, one from Kaggle (disasters: we couldn’t find relevant accessible information from somewhere else) and GitHub. We found different datasets containing useful information in relation with the SDGs. The details about these data and the links are presented in the next question.
During the wrangling process: we add data to our table (D1_1_SDG) based on different other datasets and match them based on the country, the country code, and the year. The table below shows all our 9 databases that we merge to have our final table for the analysis, as well as each variable of interest that we keep.
Table
Name
Variable
Name
Explanation
# obs before cleaning
# obs
after cleaning
In all database
code
Country code (ISO)
country
Name of the country
year
Year of the observation (2000-2022)
D1_1_
SDG
overallscore
Overall score on all 17 SDGs (the score are % of achievement of the goals determined by the UN based on several indicators)
4140 observations of 120 variables
3818 observations of 21 variables
goal1:goal17
Score on each SDG except SDG 14 (16 variables)
population
Number of people living in the country
D2_2_
Unemployment_rate
unemployment.rate
Unemployment rate (% of the population 15 years old and older)
82800 observations of 8 variables
571 observations of 5 variables
D3_0_
GDP_per_capita
GDPpercapita
GDP per capita
266 observations of 68 variables
3818 observations of 4 variables
D3_1_
Military_expenditure_
percent_GDP
MilitaryExpenditure
PercentGDP
Military expenditures in percentage of GDP
266 observations of 68 variables
3818 observations of 4 variables
D3_2_
Military_expenditure_
percent_gov_exp
MilitaryExpenditure
PercentGovExp
Military expenditures in percentage of government expenditures
266 observations of 68 variables
3818 observations of 4 variables
D4_0_
Internet_usage
internet.usage
Internet usage (% of the population)
6570 observations of 4 variables
3433 observations of 4 variables
D5_0_
Human_freedom_index
region
Part of the world, group of countries (e.g. Eastern Europe, Dub-Saharan Africa, South Asia, etc.)
3465 observations of 141 variables
3339 observations of 18 variables
hf_score
Human Freedom score = mean of personal freedom (PF) and economic freedom (EF).
pf_law
Rule of law, mean score of:
Procedural justice
Civil justice
Criminal justice
Rule of law (V-Dem)
pf_security
Security and safety, mean score of:
Homicide
Disappearances conflicts,
terrorism
pf_movement
Freedom of movement (V-Dem)
Freedom of movement (CLD)
pf_religion
Freedom of religion
Religious organizatio repression
pf_assembly
Civil society entry and exit
Freedom of assembly
Freedom to form/run political parties
Civil society repression
pf_expression
Direct attacks on the press
Media and expression (V-Dem)
Media and expression (Freedom House)
Media and expression (BTI)
Media and expression (CLD)
pf_identity
Same-sex relationships
Divorce
Inheritance rights
Female genital mutilation
pf_score
Mean of every PF component score
ef_government
Government consumption
Transfers and subsidies
Government investment
Top marginal tax rate
State ownership of assets
ef_legal
Judicial independence
Impartial courts
Protection of property rights
Military interference Integrity of the legal system Legal enforcementof contracts
Regulatory costs
Reliability of police
ef_money
Money growth
Standard deviation of inflation
Inflation: Most recent year
Freedom to own foreign currency
ef_trade
Tariffs
Regulatory trade barriers
Black-market exchange rates
Movement of capital and people
ef_regulation
Credit market regulations
Labor market regulations
Business regulations
ef_score
Mean of every EF component score
D6_0_
Disaters
continent
Continents touched by the disasters such as floods, ouragan
14644 observations of 47 variables
2435 observations
of 10 variables
total_deaths
Number of total deaths caused by the disasters
no_injured
Number of injured people
no_affected
Number of affected people
no_homeless
Number of people that lost their home and are now homeless
total_affected
Sum of people affected (sum of the variables: no_injured, no-affected, no_homeless)
total_damages
Total of infrastructure damages
D7_0_
COVID
deaths_per_million
Number of people dead due to COVID
349966 observations of 67 variables
501 observations (only between 2020-2022, before no COVID) of 6 variables
cases_per_million
Number of COVID cases
stringency
Government Response Stringency Index: composite measure based on 9 response indicators including school closures, workplace closures, and trave
D8_0_
Conflicts
ongoing
Variable coded 1 for more than 25 deaths in intrastate conflict and 0 otherwise according to UCDP/PRIO Armed Conflict Dataset 17.1.
5016 observations of 18 variables
2782 observations of 8 variables
sum_deaths
Best estimate of deaths in all categories of violence (non-state, one-sided and state-based) recorded by the Uppsala Conflict Data Program in the country based on the UCDP GED dataset (unpublished 2016 data). The location of these events is used for estimating the extent of violence.
pop_affected
Share of population affected by violence in percentage (0 to 100) measured as described above based on population data from CIESIN, the PRIO-GRID structure as well as UCDP GED.
area_affected
Area affected by conflict
maxintensity
Two different intensity levels are coded: minor armed conflicts (1) and wars (2), Takes the max intensity of conflict in the country so that it is coded 2 if there is at least one war (>=1000 deaths in intrastate conflict) during the year. Data from UCDP/PRIO Armed Conflict Dataset 17.1.
Wrangling/cleaning
To accommodate the large scale of the datasets we intended to utilize, we decided to pre-clean each of our datasets before merging them. This allowed us to simplify the process of cleaning our final dataset afterwards.
1. Dataset on SDG
This is our main dataset, that we clean in order to keep the columns containing the following inforamtion: country name, country code, year, population, overall score and the 17 SDGs and the score of each SDG exept for SDG 14 (life under water), because there was 24.1% of missing values. We couldn’t replace them, because the information was always missing for a courty across all years. Since the other SDGs such as life on earth and clean water already treat similar subjects, we decided to delete the variable. Two other goals had some missing values: SDG 1 (end poverty) with 9.04% missing values in 15 different countries and SDG 10 (reduced inequalities) with 10.2% missing values.We decide to keep both SDGs and only remove the countries with no information for the analysis, because it isn’t too much countries and that these SDGs are very important. The column population also contained some missing values, but we found out that it was for aggregate groups of countries, that we removed from the database, since we are interested in the different countries only.
# Import dataD1_0_SDG <-read.csv(here("scripts","data","SDG.csv"), sep =";")# Transform -> dataframeD1_0_SDG <-as.data.frame(D1_0_SDG)# We only want to keep certain columns: country code, country, year, population, overall SDG score and the scores on each SDGD1_0_SDG <- D1_0_SDG[,1:22]# Rename the columns to have our variablescolnames(D1_0_SDG) <-c("code", "country", "year", "population", "overallscore", "goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal14", "goal15", "goal16", "goal17")# Transform the SDG overall score into a numeric valueD1_0_SDG[["overallscore"]] <-as.double(gsub(",", ".", D1_0_SDG[["overallscore"]]))# Function to transform the SDG score into numeric valuesmakenumSDG <-function(D1_0_SDG) {for (i in1:17) { varname <-paste("goal", i, sep ="") D1_0_SDG[[varname]] <-as.double(gsub(",", ".", D1_0_SDG[[varname]])) }return(D1_0_SDG)}D1_0_SDG <-makenumSDG(D1_0_SDG)# Make sure the encoding of the country names are UTF-8D1_0_SDG$country <-stri_encode(D1_0_SDG$country, to ="UTF-8")# standardize country namesD1_0_SDG <- D1_0_SDG %>%mutate(country =countrycode(country, "country.name", "country.name", custom_match =c("T�rkiye"="Turkey")))# inspection of missing valuespropmissing <-numeric(length(D1_0_SDG))for (i in1:length(D1_0_SDG)){ proportion <-mean(is.na(D1_0_SDG[[i]])) propmissing[i] <- proportion}propmissing#> [1] 0.0000 0.0778 0.0000 0.0778 0.0000 0.0833 0.0000 0.0000 0.0000#> [10] 0.0000 0.0000 0.0000 0.0000 0.0000 0.0944 0.0000 0.0000 0.0000#> [19] 0.2278 0.0000 0.0000 0.0000# Population has some missing values, let's investigateSDG0 <- D1_0_SDG |>group_by(code) |>select(population) |>summarize(NaPop =mean(is.na(population))) |>filter(NaPop !=0)print(SDG0, n =180)#> # A tibble: 14 x 2#> code NaPop#> <chr> <dbl>#> 1 _Africa 1#> 2 _E_Euro_Asia 1#> 3 _E_S_Asia 1#> 4 _HIC 1#> 5 _LAC 1#> 6 _LIC 1#> 7 _LIC_LMIC 1#> 8 _LMIC 1#> 9 _MENA 1#> 10 _OECD 1#> 11 _Oceania 1#> 12 _SIDS 1#> 13 _UMIC 1#> 14 _World 1# Normal to have missing values because not countries but regions so we can drop these observationsD1_0_SDG <- D1_0_SDG %>%filter(!str_detect(code, "^_"))# Now there isn't any more missing values in the variable population and we will see that we have information on 166 countries:(country_number <-length(unique(D1_0_SDG$country)))#> [1] 166# Where do we have missing values in the different goal scores? SDG1 <- D1_0_SDG |>group_by(code) |>select(contains("goal")) |>summarize(Na1 =mean(is.na(goal1)),Na2 =mean(is.na(goal2)),Na3 =mean(is.na(goal3)),Na4 =mean(is.na(goal4)),Na5 =mean(is.na(goal5)),Na6 =mean(is.na(goal6)),Na7 =mean(is.na(goal7)),Na8 =mean(is.na(goal8)),Na9 =mean(is.na(goal9)),Na10 =mean(is.na(goal10)),Na11 =mean(is.na(goal11)),Na12 =mean(is.na(goal12)),Na13 =mean(is.na(goal13)),Na14 =mean(is.na(goal14)),Na15 =mean(is.na(goal15)),Na16 =mean(is.na(goal16)),Na17 =mean(is.na(goal17))) |>filter(Na1 !=0| Na2 !=0| Na3 !=0| Na4 !=0| Na5 !=0| Na6 !=0| Na7 !=0| Na8 !=0| Na9 !=0| Na10 !=0| Na11 !=0| Na12 !=0| Na13 !=0| Na14 !=0| Na15 !=0| Na16 !=0| Na17 !=0)# Print the counts for each variablekable(for (col innames(SDG1)[-1]) {print(paste(col, "count:", sum(SDG1[[col]] !=0)))})#> [1] "Na1 count: 15"#> [1] "Na2 count: 0"#> [1] "Na3 count: 0"#> [1] "Na4 count: 0"#> [1] "Na5 count: 0"#> [1] "Na6 count: 0"#> [1] "Na7 count: 0"#> [1] "Na8 count: 0"#> [1] "Na9 count: 0"#> [1] "Na10 count: 17"#> [1] "Na11 count: 0"#> [1] "Na12 count: 0"#> [1] "Na13 count: 0"#> [1] "Na14 count: 40"#> [1] "Na15 count: 0"#> [1] "Na16 count: 0"#> [1] "Na17 count: 0"
# We see that there are only missings in 3 SDG scores: 1, 10 and 14 and that when there are missings for a country, it is on all years or none. # More investigations of those 3 SDG scores. A lot of countries don't have information on those 3 SDG, should we choose to not analyse these SDGs? SDG2 <- D1_0_SDG |>group_by(code) |>select(contains("goal")) |>summarize(Na1 =mean(is.na(goal1))) |>filter(Na1 !=0)print(table(SDG2$Na1))#> #> 1 #> 15length(unique(SDG2$code))/country_number#> [1] 0.0904# there are only 9.04% missing values in 15 different countries, goal 1 being "end poverty", we decide to keep it and only remove the countries with no information for the analysisSDG3 <- D1_0_SDG |>group_by(code) |>select(contains("goal")) |>summarize(Na10 =mean(is.na(goal10))) |>filter(Na10 !=0)print(table(SDG3$Na10))#> #> 1 #> 17length(unique(SDG3$code))/country_number#> [1] 0.102# there are only 10.2% missing values in 17 different countries, goal 10 being "reduced inequalities", we decide to keep it and only remove the countries with no information for the analysisSDG4 <- D1_0_SDG |>group_by(code) |>select(contains("goal")) |>summarize(Na14 =mean(is.na(goal14))) |>filter(Na14 !=0)print(table(SDG4$Na14))#> #> 1 #> 40length(unique(SDG4$code))/country_number#> [1] 0.241# there are 24.1% missing values in 40 different countries, goal 14 being "life under water", we decide not to keep it, because other SDG such as life on earth and clean water already treat similar subjects# Delete SDG14D1_0_SDG <- D1_0_SDG %>%select(-goal14)
We will be working with different datasets and merge them based on the country code and the year. To make sure the match will work well, we standardize the name of the countries and country code using the countrycode library. In addition, we create a liste of all the country codes contained in the main database in order to filter the other databases. Finally, we complete the database to make sure all the combinations of “country, year” are in the database. The number of rows isn’t changed.
# Standardize country codeD1_0_SDG$code <-countrycode(sourcevar = D1_0_SDG$code,origin ="iso3c",destination ="iso3c",)# Create a character vector with all the different country codeslist_country <-c(unique(D1_0_SDG$code))# Create a dataframe with the list of countries and their respective codesD1_0_SDG_country_list <- D1_0_SDG %>%filter(code %in% list_country) %>%select(code, country)# remove duplicated rowsD1_0_SDG_country_list <- D1_0_SDG_country_list %>%select(code, country) %>%distinct()# Complete database to make sure there aren't couples of (year, code) missingD1_0_SDG <- D1_0_SDG |>complete(code, year)
Here are the first few lines of the cleaned dataset on SDG achievement scores:
As said, this is now our main dataset. All subsequent datasets will be merged with this dataset. Therefore, for all the following datasets, we want to make sure that we only keep data for the same countries and years as in this dataset. We have a total of 166 countries and the years range from 2000 to 2022.
2. Dataset on Unemployment rate
In this dataset, the initial step involves importing the data. Next, we ensure that the names and codes of the countries are formatted in UTF-8, preventing any discrepancies due to mismatches in country names. Following this, we modify the column names and filter the data to include only the relevant countries and years, specifically the years 2000 to 2022, covering 166 countries from our primary dataset.
Here are the first few lines of the cleaned dataset on Unemployment rate:
3. Dataset on GDP military Expenditures
We have three different databases which contain information on each countries over the years. Each year represent one variable. We want to extract three variables for our analysis: GDP per capita, military expenditures in percentage of the GDP and military expenditures in percentage of government expenditures.
After importing the data, we fill in the missing country codes using the column Indicator.Name, because we realized after some manipulations, that some of the country codes were false, but the next column contained the right ones.
fill_code <-function(data){ data <- data %>%mutate(Country.Code =ifelse(!grepl("^[A-Z]{3}$", Country.Code), Indicator.Name, Country.Code))}
We create a set of functions that we will apply to each database. First, remove the variables that we don’t need, which are the years before 2000. Second, make sure that the values are numeric and rename the year variables (because they all had an “X” before year number). Third, transform the database from wide to long, in order to match the main database. Fourth, transform the year variable into an integer variable and rearrange and rename the columns to match the ones of the other databases. Then, we apply these transformations to the three databases.
# remove the variables that we don't needremove <-function(data){ years <-seq(1960, 1999) removeyears <-paste("X", years, sep ="") data <- data[, !(names(data) %in%c("Indicator.Name", "Indicator.Code", "X", removeyears))]}# Make sure that the values are numericmakenum <-function(data) {for (i in2000:2022) { year <-paste("X", i, sep ="") data[[year]] <-as.numeric(data[[year]]) }return(data)}# Rename years from Xyear renameyear <-function(data) {for (i in2000:2022) { varname <-paste("X", i, sep ="")names(data)[names(data) == varname] <-gsub("X", "", varname) }return(data)}# Transform the database from wide to longwide2long <-function(data) { data <-pivot_longer(data, cols =-c("Country.Name", "Country.Code"), names_to ="year", values_to ="data")return(data)}# Transform the year variable into an integer variableyearint <-function(data) { data$year <-as.integer(data$year)return(data)}# Rearrange and rename the columns to match the ones of the other datasetsnameorder <-function(data) {colnames(data) <-c("country", "code", "year", "data") data <- data %>%select(c("code", "country", "year", "data"))}# One function that contains all the otherscleanwide2long <-function(data){ data <-fill_code(data) data <-remove(data) data <-makenum(data) data <-renameyear(data) data <-wide2long(data) data <-yearint(data) data <-nameorder(data)}# Apply function to three databaseGDPpercapita <-cleanwide2long(GDPpercapita)MilitaryExpenditurePercentGDP <-cleanwide2long(MilitaryExpenditurePercentGDP)MiliratyExpenditurePercentGovExp <-cleanwide2long(MiliratyExpenditurePercentGovExp)
We rename the colums with the main information, standardize the country code and remove the countries that are not in our main database. We see that all the 166 countries are there.
# Rename the data columns to have the right nameGDPpercapita <- GDPpercapita %>%rename(GDPpercapita = data)MilitaryExpenditurePercentGDP <- MilitaryExpenditurePercentGDP %>%rename(MilitaryExpenditurePercentGDP = data)MiliratyExpenditurePercentGovExp <- MiliratyExpenditurePercentGovExp %>%rename(MiliratyExpenditurePercentGovExp = data)# Standardize the country codeGDPpercapita$code <-countrycode(sourcevar = GDPpercapita$code,origin ="iso3c",destination ="iso3c",)MilitaryExpenditurePercentGDP$code <-countrycode(sourcevar = MilitaryExpenditurePercentGDP$code,origin ="iso3c",destination ="iso3c",)MiliratyExpenditurePercentGovExp$code <-countrycode(sourcevar = MiliratyExpenditurePercentGovExp$code,origin ="iso3c",destination ="iso3c",)# Remove the obervations of countries that aren't in our main dataset on SDGs: GDPpercapita <- GDPpercapita %>%filter(code %in% list_country)length(unique(GDPpercapita$code))#> [1] 166MilitaryExpenditurePercentGDP <- MilitaryExpenditurePercentGDP %>%filter(code %in% list_country)length(unique(MilitaryExpenditurePercentGDP$code))#> [1] 166MiliratyExpenditurePercentGovExp <- MiliratyExpenditurePercentGovExp %>%filter(code %in% list_country)length(unique(MiliratyExpenditurePercentGovExp$code))#> [1] 166# There are only 157 countries that are both in the main SDG dataset and in these 3 datasets. But we suspect that some of the missing countries were in the database but not rightly matchedlist_country_GDP <-c(unique(GDPpercapita$code))(missing <-setdiff(list_country, list_country_GDP))#> character(0)# 1. Bahamas was in the database but instead of the code "BHS" there is "The"# 2. "COD" "Dem. Rep."# 3. "COG" "Rep"# 4. "EGY" "Arab Rep."# 5. "GMB" "The"# 6. "IRN" "Islamic Rep."# 7. "KOR" "Rep."# 8. "VEN" "RB"# 9. "YEM" "Rep."# We remark that the code is in another column of the initial database: "Indicator.Name"# We go back to the initial database and before cleaning it we put the right codes# After rerunning the code we see that we have all our 166 countries from the initial dataset
We run a first round of investigation of the missing values and find that we have 16.4% for MiliratyExpenditurePercentGovExp, 12.9% for MilitaryExpenditurePercentGDP and 1.31% for GDPpercapita.
# What is the percentage of missing values in these 3 datasets?mean(is.na(MiliratyExpenditurePercentGovExp$MiliratyExpenditurePercentGovExp))#> [1] 0.164mean(is.na(MilitaryExpenditurePercentGDP$MilitaryExpenditurePercentGDP))#> [1] 0.129mean(is.na(GDPpercapita$GDPpercapita))#> [1] 0.0131# 16.4% for MiliratyExpenditurePercentGovExp, 12.9% for MilitaryExpenditurePercentGDP and 1.31% for GDPpercapita
GDP per capita
For GDPpercapita, only two countries (SOM and SSD) have a lot of missing values and in total 11 countries countries have missing values.
####### Investigate missing values in GDPpercapita ######GDPpercapita1 <- GDPpercapita %>%group_by(code) %>%summarize(NaGDP =mean(is.na(GDPpercapita))) %>%filter(NaGDP !=0)print(GDPpercapita1, n =180)#> # A tibble: 11 x 2#> code NaGDP#> <chr> <dbl>#> 1 AFG 0.130 #> 2 BTN 0.0435#> 3 CUB 0.0870#> 4 LBN 0.0435#> 5 SOM 0.565 #> 6 SSD 0.652 #> 7 STP 0.0435#> 8 SYR 0.0870#> 9 TKM 0.0870#> 10 VEN 0.304 #> 11 YEM 0.130# Only SOM and SSD have a lot of missings and in total 11 countries with missings
We plot the evolution of GDPpercapita avec the years for each country containing missing values and distinguish the percentage of missing values with colors.
# Create a dataframe that only have the countries with missing values and # add a column which contains the % of missings for each countryfiltered_data_GDP <- GDPpercapita %>%filter(code %in% GDPpercapita1$code)filtered_data_GDP <- filtered_data_GDP %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(GDPpercapita))) %>%ungroup()# Look at the evolution over the years for the countries that have missing valuesEvol_Missing_GDP <-ggplot(data = filtered_data_GDP) +geom_point(aes(x = year, y = GDPpercapita, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of GDP per capita over time", x ="Year", y ="GDP per capita") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "30-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4)print(Evol_Missing_GDP)# We decide not to use SSD SOM and VEN since there are more than 30% missing
For the countries with less than 30% of missing values and a linear evolution in time, we fill the missing values using linear interpolation.
####### Fill in missing values in GDPpercapita ####### Almost all have a linear evolution over time, we fill in the missing values based on the lines# AFG, BTN, CUB, STP and TKM are easy with only one linelist_code <-c("AFG", "BTN", "CUB", "STP", "TKM")for (i in list_code) {# Filter the dataset for the current country country_data <- GDPpercapita %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$GDPpercapita)# Update the original dataset with the interpolated values GDPpercapita[GDPpercapita$code == i, "GDPpercapita"] <- interpolated_data}# SYR and YEM : we fit 2 lines to fill the values# LBN: weird at the end, we don't fill the missing value for now
Military expenditures in percentage of GDP
For MilitaryExpenditurePercentGDP, 12 countries have 100% of missing values. We further investigate and keep them for now, knowing that some of these coutries may also have many missing values in the other databases when wee merge everything and will be dropped later.
We plot the evolution of MilitaryExpenditurePercentGDP along the years for each country containing missing values and distinguish the percentage of missing values with colors.
# Create a dataframe that only have the coutnries with missing values and # add a column which contains the % of missings for each countryfiltered_data_Mil1 <- MilitaryExpenditurePercentGDP %>%filter(code %in% MilitaryExpenditurePercentGDP1$code)filtered_data_Mil1 <- filtered_data_Mil1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(MilitaryExpenditurePercentGDP))) %>%ungroup()# Look at evolution over the yearsEvol_Missing_Mil1 <-ggplot(data = filtered_data_Mil1) +geom_line(aes(x = year, y = MilitaryExpenditurePercentGDP, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Military expenditure in % of GDP over time", x ="Years from 2000 to 2022", y ="GDP per capita") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =5) +theme(strip.text =element_text(size =6)) +scale_x_continuous(breaks =NULL) +scale_y_continuous(breaks =NULL)print(Evol_Missing_Mil1)
For the countries with less than 30% of missing values and a linear evolution in time, we fill the missing values using linear interpolation.
# Try to fill the missings if %missings < 30%##### Fill in missing values in MilitaryExpenditurePercentGDP ###### "AFG", "BDI", "BEN", "CAF", "CIV", "COD", "GAB", "GMB", "KAZ", "LBN", "LBR", "MNE", "MRT", "NER", "TKJ", "TTO", "ZMB"# <30% missing and linear (17)list_code <-c("AFG", "BDI", "BEN", "CAF", "CIV", "COD", "GAB", "GMB", "KAZ", "LBN", "LBR", "MNE", "MRT", "NER", "TKJ", "TTO", "ZMB")for (i in list_code) {# Filter the dataset for the current country country_data <- MilitaryExpenditurePercentGDP %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$MilitaryExpenditurePercentGDP)# Update the original dataset with the interpolated values MilitaryExpenditurePercentGDP[MilitaryExpenditurePercentGDP$code == i, "MilitaryExpenditurePercentGDP"] <- interpolated_data}# "BIH", "COG", "IRQ", "MMR", "SDN", "TCD", "TGO", "ZWE"# <30% missing but not linear (keep but we will see later) (8)# Others have too much missing (24)
Military expenditures in percentage of governement expenditures
For MilitaryExpenditurePercentGovExp, 17 countries have 100% of missing values. We further investigate and keep them for now, knowing that some of these coutries may also have many missing values in the other databases when wee merge everything and will be dropped later.
We plot the evolution of MilitaryExpenditurePercentGovExp along the years for each country containing missing values and distinguish the percentage of missing values with colors.
# Create a dataframe that only have the coutnries with missing values and add a column which contains the % of missings for each countryfiltered_data_Mil2 <- MiliratyExpenditurePercentGovExp %>%filter(code %in% MiliratyExpenditurePercentGovExp1$code)filtered_data_Mil2 <- filtered_data_Mil2 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(MiliratyExpenditurePercentGovExp))) %>%ungroup()# Look at evolution over the yearsEvol_Missing_Mil2 <-ggplot(data = filtered_data_Mil2) +geom_line(aes(x = year, y = MiliratyExpenditurePercentGovExp, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Military expenditure in % of government expenditures over time", x ="Year from 2000 to 2022", y ="GDP per capita") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =5) +theme(strip.text =element_text(size =6)) +scale_x_continuous(breaks =NULL) +scale_y_continuous(breaks =NULL)print(Evol_Missing_Mil2)
For the countries with less than 30% of missing values and a linear evolution in time, we fill the missing values using linear interpolation.
# Try to fill the missings if %missings < 30%##### Fill in missing values in MilitaryExpenditurePercentGovExp ###### "AFG", "ARM", BEN", "BIH", "BLR", COG", "ECU", "GAB", "GMB", "KAZ", "LBN", "LBR", "MNE", "MWI", "NER", "TTO", "UKR", "ZMB" <30% missing and linear (18)list_code <-c("AFG", "ARM", "BEN", "BIH", "BLR", "COG", "ECU", "GAB", "GMB", "KAZ", "LBN", "LBR", "MNE", "MWI", "NER", "TTO", "UKR", "ZMB")for (i in list_code) {# Filter the dataset for the current country country_data <- MiliratyExpenditurePercentGovExp %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$MiliratyExpenditurePercentGovExp)# Update the original dataset with the interpolated values MiliratyExpenditurePercentGovExp[MiliratyExpenditurePercentGovExp$code == i, "MiliratyExpenditurePercentGovExp"] <- interpolated_data}# "BDI", "IRQ"# 2 lines (2)# "CAF", "MMR", "SDN", "TCD", "TGO", "TJK" # <30% missing but not linear (keep but we will see later) (6)# Others have too much missing (31) -> very much maybe we will have to drop this variable for our analysis
We now look again at the percentage of missing values for the trhee databases: 14.49% for MiliratyExpenditurePercentGovExp, 11.6% for MilitaryExpenditurePercentGDP and 1.07% for GDPpercapita
# And now, What is the percentage of missing values in these 3 datasets?mean(is.na(MiliratyExpenditurePercentGovExp$MiliratyExpenditurePercentGovExp))#> [1] 0.149mean(is.na(MilitaryExpenditurePercentGDP$MilitaryExpenditurePercentGDP))#> [1] 0.116mean(is.na(GDPpercapita$GDPpercapita))#> [1] 0.0107# Standardize names for mergeD3_1_GDP_per_capita <- GDPpercapitaD3_2_Military_Expenditure_Percent_GDP <- MilitaryExpenditurePercentGDPD3_3_Miliraty_Expenditure_Percent_Gov_Exp <- MiliratyExpenditurePercentGovExp
Here are the first few lines of the cleaned dataset of GDP per capita:
Here are the first few lines of the cleaned dataset of military expenditures in percentage of GDP:
Here are the first few lines of the cleaned dataset of military expenditures in percentage of government expenditures:
4. Dataset on internet usage
To prepare the dataset on internet usage in the world to be merge with the other data, we first, import the data. Then, we keep only the year that we are interested in (2000 to 2022). We also rename the column and keep only the country that match the list of the countries in the main dataset on the SDG.
Here are the first few lines of the cleaned dataset of internet usage:
5. Dataset on human freedom index
After importing the data from the CATO Institute website, we noticed that even if the file was called “Human Freedom Index 2022”, the available observations were only going from 2000 up to 2020. We have decided first to modify it in order to match our other datasets, by renaming/encoding/standardizing the columns containing the country names.
data <-read.csv(here("scripts", "data", "human-freedom-index-2022.csv"))#data in tibble datatibble <-tibble(data)# Rename the column countries into country to match the other datbasesnames(datatibble)[names(datatibble) =="countries"] <-"country"# Make sure the encoding of the country names are UTF-8datatibble$country <-iconv(datatibble$country, to ="UTF-8", sub ="byte")# standardize country namesdatatibble <- datatibble %>%mutate(country =countrycode(country, "country.name", "country.name"))
Once done, we could verify which countries were or were not present between these observations and our main SDG dataset. We have decided to keep the ones that were matching between the two datasets.
# Merge by country namedatatibble <- datatibble %>%left_join(D1_0_SDG_country_list, by ="country")datatibble <- datatibble %>%filter(code %in% list_country)(length(unique(datatibble$code)))#> [1] 159# See which ones are missinglist_country_free <-c(unique(datatibble$code))(missing <-setdiff(list_country, list_country_free))#> [1] "AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB"# Turkey was missing but present in the initial database (it was a problem when stadardizing the country names of D1_0SDG_country_list that we corrected) and the other missing countries are:"AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB" D5_0_Human_freedom_index <- datatibble
Then, we noticed that there were a lot of columns that were not important for us, as we had 141 variables taken into account. So we have decided to keep the ones that refers to the countries informations (such as code, year, ..) and their human freedom scores per category (pf for personnal freedom, ef for economical freedom).
# erasing useless columns to keep only the general ones. D5_0_Human_freedom_index <-select(D5_0_Human_freedom_index, year, country, region, hf_score, pf_rol, pf_ss, pf_movement, pf_religion, pf_assembly, pf_expression, pf_identity, pf_score, ef_government, ef_legal, ef_money, ef_trade, ef_regulation, ef_score, code)D5_0_Human_freedom_index <- D5_0_Human_freedom_index %>%rename(pf_law =names(D5_0_Human_freedom_index)[5], # Renames the 5th column to "pf_law"pf_security =names(D5_0_Human_freedom_index)[6] # Renames the 6th column to "pf_security" )
After renaming the columns pf_law/security for comprehension purpose, we have investigated how are distributed the NA values among the countries and the variables. After having found the percentages of missing values per country and variable, heatmaps revealed themself to be a great tool for visualizing datas.
Then, for having a better understanding of the situation, we ordered the countries having at least 1 variable containing 50% and more of missing values
na_long <- na_long %>%group_by(country) %>%mutate(Count_NA_50_100 =sum(NA_Percentage >=50& NA_Percentage <=100, na.rm =TRUE)) %>%ungroup() %>%arrange(desc(Count_NA_50_100))heatmap_ordered_all <-ggplot(na_long, aes(x =reorder(country, -Count_NA_50_100), y = Variable)) +geom_tile(aes(fill = NA_Percentage), colour ="white") +scale_fill_gradient(low ="white", high ="red") +theme_minimal() +labs(title ="Heatmap of NA Percentages per Country and Variable",x ="Countries",y ="Variables",fill ="NA Percentage" ) +theme(axis.text.x =element_blank(), # Hide x-axis labelsaxis.text.y =element_text(size =9) )print(heatmap_ordered_all)
We notice that only some countries look to contain at least 50% of missing values and in addition that most of the missing values are concerning the EF variables (Economic Freedom). Now, we tried to produce another heatmap only containing the ordered countries, and also counting for each one of these country the number of variables with at least 50% of NAs.
We conclude here that 13 countries were concerned by our selection of 50% and more of missing values. When discussing between us, we came to the conclusion that among these 13 countries, a great part of them were not going to be selected because they had a lot of missing values in our main dataset too. Therefore, we have decided to merge this data with the other datasets and finish the cleaning after.
Here are the first few lines of the partialy cleaned dataset on Human Freedom Index scores:
6. Dataset on Disasters
For this dataset concerning the Disasters we imported the data from Kaggle as we couldn’t find the original dataset that is private coming from the EOSDIS SYSTEM, an interactive interface for browsing full-resolution, global, daily satellite images from NASA. Once we made sure that our file called “Disasters” was convert into a data frame, we selected some specific columns that we where interested in.
Because we knew that our file showed all the disasters in each country over the years (between 1970-2021) and that we wanted to focus on a specific period, we filtered our data to show the years between 2000 and 2022. Then we rearranged our data, changing the data types of all the columns and their names in order to match our other datasets.
# Rearrange the columns, changed the type of data, renamed the columnsRearanged_Disasters <- Disasters %>%filter(Year >=2000& Year <=2022) %>%mutate(code =as.character(ISO),country =as.character(Country),year =as.integer(Year),continent =as.character(Continent),disaster.subgroup =as.character(Disaster.Subgroup),disaster.type =as.character(Disaster.Type),location =as.character(Location),total.deaths =as.numeric(Total.Deaths),no.injured =as.numeric(No.Injured),no.affected =as.numeric(No.Affected),no.homeless =as.numeric(No.Homeless),total.affected =as.numeric(Total.Affected),total.damages =as.numeric(Total.Damages...000.US..) )
We then grouped the data by “year”, “code”, “country” and “continent” and summarize the data. Here you can see that we re-selected specific columns as we saw that our first pre-selection was still too wide and some variables as the disaster.subgroup and disaster.type weren’t pertinent.We arranged the columns based on “code,” “country,” “year,” and “continent” to match the other datasets.
Finally we filtered our disasters data to keep only the countries that are present in our main dataset. We analysed the missing countries and identified three countries (BHR, BRN, MLT) that are unexpectedly missing.
D6_0_Disasters <- D6_0_Disasters %>%filter(code %in% list_country)length(unique(D6_0_Disasters$code))#> [1] 163# Here we see which countries are missinglist_country_disasters <-c(unique(D6_0_Disasters$code))(missing <-c(missing,setdiff(list_country, list_country_disasters)))#> [1] "AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB" "BHR" "BRN" "MLT"
Here are the first few lines of the cleaned dataset on Disasters:
7. Dataset on COVID
This dataset contains information on the COVID19 pandemic between 2020 and 2022. The observation are by year, month, day. After importing the database, we transform the date in format YYYY-MM-DD in order to only keep the year.
#Import dataCOVID <-read.csv(here("scripts","data","COVID.csv"))# Only keep the variables that we are interested inCOVID <- COVID[,c("iso_code", "location", "date", "new_cases_per_million", "new_deaths_per_million", "stringency_index")]# Transform the dates ("YYYY-MM-DD") into years ("YYYY") and integersCOVID$date <-as.integer(year(COVID$date))
We perform a first round of investigation of the missing values before aggregating the values by year. We begin with the variables “cases per million” and “deaths per million”: seeing that for each country, we have either only missing values, either a very low percentage of missing values (~1%), we can compute the sum over each year and ignore the missing values without altering the data. Indeed, where al the values are missing, the computation will return a NA. We then look at the “stringency” variable and we have 3 scenarios:
~20% missings: we ignore missing values when computing the mean to have an idea of stringency each year (because we compute the mean stringency over the year, if some days are missing, it is not a problem, it can not evoluate that fast).
all are missing : we can ignore the missing values when computing the mean, because it will still return a missing value
almost all are missing: here the mean doesn’t make sense -> we will replace the values by NAs to be coherent. The countries with this issues are: ERI, GUM, PRI and VIR. We verify if they are in our main dataset and since none of these countries are, we can ignore the issue, the lines will be remove later anyway.
We aggregate the observations of all days of a year in one observation per country using the mean.
# Investigate missing values before aggregatingCOVID1 <- COVID %>%group_by(iso_code) %>%summarize(NaCOVID =round(mean(is.na(new_cases_per_million)),3)) %>%filter(NaCOVID !=0)print(table(COVID1$NaCOVID))#> #> 0.001 0.002 0.003 0.004 0.012 0.109 1 #> 33 6 2 5 1 1 9COVID2 <- COVID %>%group_by(iso_code) %>%summarize(NaCOVID =round(mean(is.na(new_deaths_per_million)),3)) %>%filter(NaCOVID !=0)print(table(COVID2$NaCOVID))#> #> 0.001 0.002 0.004 0.11 1 #> 32 1 2 1 9# We see that for each country, we have either only missing values, either a very low % of missing ~1% -> we can compute the sum over each year and ignore the missing values withoutaltering the dataCOVID3 <- COVID %>%group_by(iso_code) %>%summarize(NaCOVID =round(mean(is.na(stringency_index)), 3)) %>%filter(NaCOVID !=0)print(table(COVID3$NaCOVID))#> #> 0.13 0.186 0.198 0.21 0.986 1 #> 1 1 1 178 4 70# Here we have 3 scenarios: # 1) ~20% missings -> ok to ignore missings when computing the mean to have an idea of stringency each year# 2) all missings -> ok to ignore missings when computing the mean because it will still return a missing value# 3) almost all are missing: here the mean doesn't make sense -> we will replace the values by NAs to be coherent. The countries with this issues are: ERI, GUM, PRI and VIR, we want to see if these countries are in our main datasetissue_list <-c("ERI", "GUM", "PRI", "VIR")is.element(issue_list, list_country)#> [1] FALSE FALSE FALSE FALSE# Since non of these countries are in the main SDG database, we can ignore the issue, the lines will be remove later anyway# Aggregate the observation of all days of a year in one observation per countryCOVID <- COVID %>%group_by(location, date) %>%mutate(cases_per_million =sum(new_cases_per_million, na.rm =TRUE),deaths_per_million =sum(new_deaths_per_million, na.rm =TRUE),stringency =mean(stringency_index, na.rm =TRUE) )%>%ungroup()
Now that all the variables of interest are aggregated by year, we remove all the variables that we don’t need and rename all the remaining variables to match the main dataset.
# Only have 1 obs per country per yearCOVID <- COVID %>%group_by(location, date) %>%distinct(date, .keep_all =TRUE) %>%ungroup()# Remove the variable that have the information for every day and only keep those by yearCOVID <- COVID %>%select(-c(new_cases_per_million, new_deaths_per_million, stringency_index))# Rename the variablescolnames(COVID) <-c("code", "country", "year", "cases_per_million", "deaths_per_million", "stringency")
We remove the years that exceed 2022, we make sure that the country codes are all iso codes with 3 letters (we observe that sometimes they are preceded by “OWID_”) and we standardize the country codes.
# Remove the years after 2022 to match our main database COVID <- COVID[COVID$year <=2022, ]# Make sure the country codes are all iso codes with 3 letters (we observe that sometimes they are preceded by "OWID_")COVID$code <-gsub("OWID_", "", COVID$code)# Standardize the country codeCOVID$code <-countrycode(sourcevar = COVID$code,origin ="iso3c",destination ="iso3c",)
We remove the observations of countries that aren’t in our main dataset on SDGs and find that all the 166 countries that we have in the main SDG dataset are also in this one.
# Remove the observations of countries that aren't in our main dataset on SDGs: COVID <- COVID %>%filter(code %in% list_country)length(unique(COVID$code))#> [1] 166# All the 166 countries that we have in the main SDG dataset are also in this one.
We perform a second round of missing values investigation and find out that there are no missing values except for the stringency, where there are 4.19%. Either all values are missing for one country, or 50% are missing, so these 7 countries won’t be included when analyzing the effect of stringency on the SDG scores.
##### Investigation of the missing values #####mean(is.na(COVID$cases_per_million))#> [1] 0mean(is.na(COVID$deaths_per_million))#> [1] 0mean(is.na(COVID$stringency))#> [1] 0.0419# No missing values except in for the stringency, where there are 4.19% COVID4 <- COVID %>%group_by(code) %>%summarize(NaCOVID =mean(is.na(stringency))) %>%filter(NaCOVID !=0)print(COVID4, n =300)#> # A tibble: 7 x 2#> code NaCOVID#> <chr> <dbl>#> 1 ARM 1 #> 2 COM 1 #> 3 MDV 1 #> 4 MKD 1 #> 5 MNE 1 #> 6 NAM 0.5#> 7 STP 1# Either all values are missing for one country, or 50% are missing, so these 7 countries # won't be included when analysing the effect of stringencyD7_0_COVID <- COVID
Here are the first few lines of the cleaned dataset on COVID19:
8. Dataset on Conflicts
For our conflicts dataset, we imported the data from “The World Banck” data catalog. Once we made sure that our file called “Disasters” was convert into a data frame, we selected some specific columns that we where interested in.
Our file showed all the Conflicts and consequences per country over the years (between 2000-2016). We couldn’t find a better and more complete dataset, As we consider conflicts as events, we will only take into account results between 2000 and 2016. Then we rearranged our data, changing the data types of all the columns and their names in order to match our other datasets. We grouped the data by ” year”, “country”, re-selected some variables and summarize the data.
Rearanged_Conflicts <- Conflicts %>%filter(year >=2000& year <=2022)%>%mutate(ongoing =as.integer(ongoing),country =as.character(country),year =as.integer(year),gwsum_bestdeaths =as.numeric(gwsum_bestdeaths),pop_affected =as.numeric(pop_affected),area_affected =as.numeric(area_affected),maxintensity =as.numeric(maxintensity), )# Group the data by "year", "country" and summarize the dataConflicts <- Rearanged_Conflicts %>%group_by(year, country) %>%summarize(ongoing =sum (ongoing, na.rm =TRUE),sum_deaths =sum(gwsum_bestdeaths, na.rm =TRUE),pop_affected =sum(pop_affected, na.rm =TRUE),area_affected =sum(area_affected, na.rm =TRUE),maxintensity =sum(maxintensity, na.rm =TRUE), )
After we Selected specific columns from the summarized data and arrange the data by our specified columns. To make our dataset compatible with the main one and let the merging face succeed, we dd some adjustment concerning the country names’ to ensure the compatibility. Then we standardize and merge by country names to finally rearrange the data to retain only the countries present in our main dataset. Note that in the end we can see that only one country is missing that wasn’t in the initial conflicts database: BLR
conflicts <- Conflicts %>%select(country, year, ongoing, sum_deaths, pop_affected, area_affected, maxintensity) %>%arrange(country, year)conflicts$country <-iconv(conflicts$country, to ="UTF-8", sub ="byte")conflicts <- conflicts %>%mutate(country =countrycode(country, "country.name", "country.name"))conflicts <- conflicts %>%left_join(D1_0_SDG_country_list, by ="country")conflicts <- conflicts %>%select(code, country, year, ongoing, sum_deaths, pop_affected, area_affected, maxintensity) %>%arrange(code, country, year)D8_0_Conflicts <- conflicts %>%filter(code %in% list_country)(length(unique(conflicts$code)))#> [1] 166# See which countries are missinglist_country_conflicts <-c(unique(conflicts$code))(missing <-c(missing, setdiff(list_country, list_country_conflicts)))#> [1] "AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB" "BHR" "BRN" "MLT"#> [11] "BLR"
Here are the first few lines of the cleaned dataset on Conflicts:
Merge data
By merging our eight pre-cleaned datasets, we create a final database.
# merge D1_0_SDG with D2_1_Unemployment_rate D2_1_Unemployment_rate$country <-NULLmerge_1_2 <- D1_0_SDG |>left_join(D2_1_Unemployment_rate, join_by(code, year))# merge merge_1_2 with D3_1_GDP_per_capita, D3_2_Military_Expenditure_Percent_GDP and D3_3_Miliraty_Expenditure_Percent_Gov_ExpD3_1_GDP_per_capita$country <-NULLmerge_12_3 <- merge_1_2 |>left_join(D3_1_GDP_per_capita, join_by(code, year))D3_2_Military_Expenditure_Percent_GDP$country <-NULLmerge_12_3 <- merge_12_3 |>left_join(D3_2_Military_Expenditure_Percent_GDP, join_by(code, year)) D3_3_Miliraty_Expenditure_Percent_Gov_Exp$country <-NULLmerge_12_3 <- merge_12_3 |>left_join(D3_3_Miliraty_Expenditure_Percent_Gov_Exp, join_by(code, year)) # merge merge_12_3 with D4_0_Internet_usage D4_0_Internet_usage$country <-NULLmerge_123_4 <- merge_12_3 |>left_join(D4_0_Internet_usage, join_by(code, year)) # merge merge_123_4 with D5_0_Human_freedom_indexD5_0_Human_freedom_index$country <-NULLmerge_1234_5 <- merge_123_4 |>left_join(D5_0_Human_freedom_index, join_by(code, year)) # merge merge_1234_5 with D_6_0_DisastersD6_0_Disasters$country <-NULLmerge_12345_6 <- merge_1234_5 |>left_join(D6_0_Disasters, join_by(code, year)) # merge merge_12345_6 with D7_0_COVIDD7_0_COVID$country <-NULLD7_0_COVID <- D7_0_COVID |>distinct(code, year, .keep_all =TRUE)merge_123456_7 <- merge_12345_6 |>left_join(D7_0_COVID, join_by(code, year)) # merge merge_123456_7 with D8_0_ConflictsD8_0_Conflicts$country <-NULLall_Merge <- merge_123456_7 |>left_join(D8_0_Conflicts, join_by(code, year)) # Filter to delete the countries that were missing from some of our databasesall_Merge <- all_Merge %>%filter(!code %in% missing)
Cleaning of the final database
We replace the NAs of the COVID columns by 0 (because we don’t have real missing, only introduced by merging for the years before COVID).
# Replace the NAs of the COVID columns by 0 (because we don't have real missing, only introduced by merging for the years before COVID)all_Merge <- all_Merge %>%mutate(cases_per_million =ifelse(is.na(cases_per_million), 0, cases_per_million),deaths_per_million =ifelse(is.na(deaths_per_million), 0, deaths_per_million),stringency =ifelse(is.na(stringency), 0, stringency) )
Since we took the information on the continent and region from databases that are not the main one, we complete these inforamtion for the whole final dataset.
# Complete the values of continent and regionall_Merge <- all_Merge %>%group_by(country) %>%mutate(continent =ifelse(is.na(continent), first(na.omit(continent)), continent)) %>%ungroup()all_Merge <- all_Merge %>%group_by(country) %>%mutate(region =ifelse(is.na(region), first(na.omit(region)), region)) %>%ungroup()
We order the database, beginning by the information on the country, the year, the continent and the region.
Here are the first few lines of the final dataset:
Final structure of our merged database: each country of the 166 countries from D1_1_SDG are observed each year from 2000 to 2022, thus each row has a key composed of (code, year) that uniquely identifies an observation. The other columns are the variables listed above. Due to some countries having a lot of missing information we will have to eliminate some of them, but we will still have more than 2000 rows in our database.
Treatment of missing values
We load our final database and subset it according to the data that we will need in order to answer the different questions. This will help us dealing with the missing values.
For question 1, we only keep the years until 2020, because most of the explanatory variables that we want to use (those coming from the human freedom index) only have values until 2020.
# subset of data# for question 1: factors (only until 2020 because no information for freedom index after)data_question1 <- all_Merge %>%filter(year<=2020) %>%select(-c(total_deaths, no_injured, no_affected, no_homeless, total_affected, total_damages, cases_per_million, deaths_per_million, stringency, ongoing, sum_deaths, pop_affected, area_affected, maxintensity))
For question 2 and 4, we use the main data from the SDG database.
# for question 2 and 4: time and relationship between SDGsdata_question24 <- all_Merge %>%select(c(code, year, country, continent, region, overallscore, goal1, goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal10, goal11, goal12, goal13, goal15, goal16, goal17))
For question 3, we create 3 distinct databases according to the different type of event that we wwill analyse: disasters, COVID19 and conflicts. For the disasters, we only keep the years until 2021, because after this date, we don’t have data. For the conflicts, we only keep the years until 2016, because after this date, we don’t have data.
# for question 3: events# Disasters (only until 2021 because no information for disasters after)data_question3_1 <- all_Merge %>%filter(year<=2021) %>%select(c(code, year, country, continent, region, overallscore, goal1, goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal10, goal11, goal12, goal13, goal15, goal16, goal7, total_deaths, no_injured, no_affected, no_homeless, total_affected, total_damages))# COVIDdata_question3_2 <- all_Merge %>%select(c(code, year, country, continent, region, overallscore, goal1, goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal10, goal11, goal12, goal13, goal15, goal16, goal7, cases_per_million, deaths_per_million, stringency))# Conflicts (only until 2016 because no information for conflicts after)data_question3_3 <- all_Merge %>%filter(year<=2016) %>%select(c(code, year, country, continent, region, overallscore, goal1, goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal10, goal11, goal12, goal13, goal15, goal16, goal7, ongoing, sum_deaths, pop_affected, area_affected, maxintensity))##### Which countries have many missing observations over the different variables of the different subsets?
Data for question 1
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. We decide to remove the countries that have more than 50 missing values.
Here is the dataframe that allows us to see the countries that have missing values, how many and for which variables, when there are more than 50 in total.
code
unemployment.rate
GDPpercapita
MilitaryExpenditurePercentGDP
MiliratyExpenditurePercentGovExp
internet_usage
hf_score
pf_law
pf_security
pf_movement
pf_religion
pf_assembly
pf_expression
pf_identity
pf_score
ef_government
ef_legal
ef_money
ef_trade
ef_regulation
ef_score
num_missing
BHS
0
0
21
21
0
0
14
0
0
0
0
0
0
0
0
0
0
0
0
0
56
BTN
0
0
21
21
0
13
0
0
0
0
0
0
0
13
0
0
13
13
10
13
117
COM
0
0
21
21
3
19
0
0
0
0
0
0
0
19
19
19
19
19
19
19
197
CPV
0
0
0
0
0
10
0
0
0
0
0
0
0
10
0
0
10
10
10
10
60
DJI
0
0
15
12
0
19
0
0
0
0
0
0
0
19
19
19
19
19
19
19
179
GIN
0
0
7
7
0
13
0
0
0
0
0
0
0
13
0
0
13
13
11
13
90
GMB
0
0
0
0
0
10
0
0
0
0
0
0
0
10
2
0
10
10
10
10
62
IRQ
0
0
4
4
2
16
0
0
0
0
0
0
0
16
3
0
16
16
16
16
109
KHM
0
0
0
0
3
10
0
0
0
0
0
0
0
10
0
0
10
10
10
10
63
LAO
0
0
7
7
0
14
0
0
0
0
0
0
0
14
0
0
14
14
13
14
97
LBN
0
0
0
0
0
10
0
0
0
0
0
0
0
10
0
0
10
10
10
10
60
LBR
0
0
0
0
2
14
0
0
0
0
0
0
0
14
0
0
14
14
10
14
82
QAT
0
0
12
12
0
10
0
0
0
0
0
0
0
10
7
0
10
10
10
10
91
SAU
0
0
0
0
0
10
0
0
0
0
0
0
0
10
0
0
10
10
10
10
60
SDN
0
0
5
5
9
16
0
0
0
0
0
0
0
16
0
0
16
16
16
16
115
SOM
0
13
21
13
4
19
0
0
0
0
0
0
0
19
19
19
19
19
19
19
203
SUR
0
0
21
21
0
10
0
0
0
0
0
0
10
10
5
0
10
10
10
10
117
SWZ
0
0
0
21
3
10
0
0
0
0
0
0
0
10
0
0
10
10
10
10
84
TJK
0
0
4
5
3
10
0
0
0
0
0
0
0
10
0
0
10
10
10
10
72
YEM
0
1
5
5
3
10
0
0
0
0
0
0
0
10
10
0
10
10
10
10
84
Now, looking at the remaining countries that have missing values and there number accross all variables, we decide to remove MilitaryExpenditurePercentGovExp, because it has too many missing values and it contains similar information to MilitaryExpenditurePercentGDP.
see_missing1_2 <- data_question1 %>%group_by(code) %>%summarise(across(-c(X, year, country, continent, region, population, overallscore, goal1, goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal10, goal11, goal12, goal13, goal15, goal16, goal17),~sum(is.na(.))) %>%mutate(num_missing =rowSums(across(everything()))) %>%filter(num_missing >0))# Delete MilitaryExpenditurePercentGovExp because it has too many missing values and contains similar information to MilitaryExpenditurePercentGDPdata_question1 <- data_question1 %>%select(-MiliratyExpenditurePercentGovExp)
Here is the dataframe that allows us to see the countries that have missing values, how many and for which variables, after remoying the countries with more than 50.
code
unemployment.rate
GDPpercapita
MilitaryExpenditurePercentGDP
MiliratyExpenditurePercentGovExp
internet_usage
hf_score
pf_law
pf_security
pf_movement
pf_religion
pf_assembly
pf_expression
pf_identity
pf_score
ef_government
ef_legal
ef_money
ef_trade
ef_regulation
ef_score
num_missing
AGO
0
0
0
0
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
30
ARE
0
0
6
6
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
12
ARM
0
0
0
0
0
4
0
0
0
0
0
0
0
4
0
0
4
4
3
4
23
AUS
0
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
AZE
0
0
0
0
2
4
0
0
0
0
0
0
0
4
0
0
4
4
0
4
22
BDI
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
BFA
0
0
0
0
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
30
BIH
0
0
2
0
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
32
BLZ
0
0
0
0
3
0
13
0
0
0
0
0
0
0
0
0
0
0
0
0
16
BRB
0
0
21
21
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
45
CAF
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
CIV
0
0
0
21
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
22
COD
0
0
0
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
21
COG
0
0
6
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
9
CRI
0
0
21
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
42
ETH
0
0
0
0
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
30
FJI
0
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
GEO
0
0
0
0
0
2
0
0
0
0
0
0
0
2
0
0
3
2
0
2
11
GUY
0
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
HTI
0
0
13
13
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
26
ISL
0
0
21
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
42
JAM
0
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
JOR
0
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
KAZ
0
0
0
0
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
30
KGZ
0
0
0
0
0
5
0
0
0
0
0
0
0
5
1
0
5
5
5
5
31
LKA
0
0
0
0
6
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
6
LSO
0
0
0
0
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
30
MDA
0
0
0
0
3
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
33
MDG
0
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
MKD
0
0
0
0
0
3
0
0
0
0
0
0
0
3
0
0
3
3
0
3
15
MMR
0
0
6
6
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
13
MNE
0
0
0
0
4
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
34
MNG
0
0
0
0
4
4
0
0
0
0
0
0
0
4
0
0
4
4
0
4
24
MOZ
0
0
0
0
0
3
0
0
0
0
0
0
0
3
0
0
3
3
0
3
15
MRT
0
0
0
7
0
5
0
0
0
0
0
0
0
5
0
0
5
5
5
5
37
MWI
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
NER
0
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
PAK
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
PAN
0
0
21
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
42
PNG
0
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
RWA
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
2
SRB
0
0
0
0
4
5
0
0
0
0
0
0
0
5
2
0
5
5
5
5
36
SYR
0
0
10
10
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
20
TCD
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
TGO
0
0
5
5
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
10
TTO
0
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
USA
0
0
0
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
21
VEN
0
5
0
21
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
29
VNM
0
0
5
5
0
2
0
0
0
0
0
0
0
2
0
0
3
2
0
2
21
ZWE
0
0
3
8
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
11
GDP per capita
Only Venezuela has missing values that we can not fill, so we delete the country.
# GDPpercapitaquestion1_missing_GDP <- data_question1 %>%group_by(code) %>%summarize(NaGDPpercapita =mean(is.na(GDPpercapita)))%>%filter(NaGDPpercapita !=0)# Only VEN, we can't fill the missing, we delete VENdata_question1 <- data_question1 %>%filter(code!="VEN")# Update list countries deletedlist_country_deleted <-c(list_country_deleted, "VEN")
Military expenditure in % of GDP
To begin with, we delete the countries with more than 30% missing values.
# Military expenditure in % of GDPquestion1_missing_Military <- data_question1 %>%group_by(code) %>%summarize(NaMilitary =mean(is.na(MilitaryExpenditurePercentGDP)))%>%filter(NaMilitary !=0)# Remove the countries with more than 30% missingdata_question1 <- data_question1 %>%filter(code!="BRB"& code!="CRI"& code!="HTI"& code!="ISL"& code!="PAN"& code!="SYR") # Update list countries deletedlist_country_deleted <-c(list_country_deleted, "BRB", "CRI", "HTI", "ISL", "PAN", "SYR")
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values, where there are less than 30% missing using the median by region.
# Create a dataframe that only have the countries with missing values and # add a column which contains the % of missings for each countryquestion1_missing_Military <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(MilitaryExpenditurePercentGDP))) %>%ungroup() %>%group_by(region) %>%filter(sum(PercentageMissing, na.rm =TRUE) >0)# See the distribution of the missings per regionFreq_Missing_Military <-ggplot(data = question1_missing_Military) +geom_histogram(aes(x = MilitaryExpenditurePercentGDP, fill =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%"))),bins =30) +labs(title ="Distribution of Military expenditures in % of GDP", x ="Military expenditures in % of GDP", y ="Frequency") +scale_fill_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red","30-100%"="black"), labels =c("0-10%", "10-20%", "20-30%","30-100%")) +guides(fill =guide_legend(title ="% missings")) +facet_wrap(~ region, nrow =3)print(Freq_Missing_Military)# All are skewed distributions, we decide to replace the missing values where there are less than 30% missing by the median by regiondata_question1 <- data_question1 %>%group_by(code) %>%mutate(PercentageMissingByCode =mean(is.na(MilitaryExpenditurePercentGDP)) ) %>%ungroup() %>%# Remove grouping temporarilygroup_by(region) %>%mutate(MedianByRegion =median(MilitaryExpenditurePercentGDP, na.rm =TRUE),MilitaryExpenditurePercentGDP =ifelse( PercentageMissingByCode <0.3&!is.na(MilitaryExpenditurePercentGDP), MilitaryExpenditurePercentGDP,ifelse(PercentageMissingByCode <0.3, MedianByRegion, MilitaryExpenditurePercentGDP) ) ) %>%select(-PercentageMissingByCode, -MedianByRegion)
Internet usage
There are only low percentage of missing values.
# Internet usagequestion1_missing_Internet <- data_question1 %>%group_by(code) %>%summarize(NaInternet =mean(is.na(internet_usage)))%>%filter(NaInternet !=0)# Only low % of missing
We look at the evolution of the variable over time. We fill the missing values with linear interpolation, because all evolutions are in an increasing way and are almost straight lines, except for CIV that we delete.
# Create a dataframe that only have the countries with missing values and # add a column which contains the % of missings for each countryquestion1_missing_Internet <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(internet_usage))) %>%filter(code %in% question1_missing_Internet$code)# Look at the evolution over the years for the countries that have missing valuesEvol_Missing_Internet <-ggplot(data = question1_missing_Internet) +geom_point(aes(x = year, y = internet_usage, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of internet usage over time", x ="Years from 2000 to 2022", y ="Internet usage") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +scale_x_continuous(breaks=NULL)+facet_wrap(~ code, nrow =4)print(Evol_Missing_Internet)# Fill with linear interpolation, because all evolution are in an increasing way and are almost straight lines, except for CIVlist_code <-setdiff(unique(question1_missing_Internet$code), "CIV")for (i in list_code) {# Filter the dataset for the current country country_data <- data_question1 %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$internet_usage)# Update the original dataset with the interpolated values data_question1[data_question1$code == i, "internet_usage"] <- interpolated_data}# Delete country CIVdata_question1 <- data_question1 %>%filter(code!="CIV")# Update list countries deletedlist_country_deleted <-c(list_country_deleted, "CIV")
Human freedom index
First, we remove hf_score, pf_score and ef_score, because there are many missing values and since these variables summarize the other ones, deleting the will not make us loose information.
# Human Freedom Index# Remove hf_score, pf_score and ef_score because many missing and since these variables are summary of other ones, deleting the mwon't make us loose informationdata_question1 <- data_question1 %>%select(-c(hf_score, pf_score, ef_score))# pf_law has (many) missing only for one country:BLZ, we remove it data_question1 <- data_question1 %>%filter(code!="BLZ")# Update list countries deletedlist_country_deleted <-c(list_country_deleted, "BLZ")
Economic freedom: government
Only KGZ and SRB have missing values, we plot the values over time and fill in the missing values by the year before, since there are only one and two missing values.
# ef_government: KGZ and SRB have missing values -> plot# KGZEvol_Missing_ef_gov <- data_question1 %>%group_by(code) %>%filter(code=="KGZ")ggplot(Evol_Missing_ef_gov, aes(x = year, y = ef_government)) +geom_point() +labs(title ="Evolution of economic freedom: government over time in KGZ", x ="Years", y ="ef_gov")# Only one missing, in 2000, replace by the value of 2001# SRBEvol_Missing_ef_gov <- data_question1 %>%group_by(code) %>%filter(code=="SRB")ggplot(Evol_Missing_ef_gov, aes(x = year, y = ef_government)) +geom_point() +labs(title ="Evolution of economic freedom: government over time in SRB", x ="Years", y ="ef_gov")# Only 2 missing, replace by next valuedata_question1 <- data_question1 %>%mutate(ef_government =ifelse(code =="KGZ"& year ==2000&is.na(ef_government), ef_government[which(code =="KGZ"& year ==2001)], ef_government))data_question1 <- data_question1 %>%mutate(ef_government =ifelse(code =="SRB"& year ==2000&is.na(ef_government), ef_government[which(code =="SRB"& year ==2002)], ef_government))data_question1 <- data_question1 %>%mutate(ef_government =ifelse(code =="SRB"& year ==2001&is.na(ef_government), ef_government[which(code =="SRB"& year ==2002)], ef_government))
Economic freedom: money
18 countries have missing values, but the percentage of missing values is always below 25%.
We look at the evolution of the variable over time. For the countries where this evolution is linear, we fill in the missing values using linear interpolation.
# Create a dataframe that only have the countries with missing values and # add a column which contains the % of missings for each countryquestion1_missing_ef_money <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_money))) %>%filter(code %in% question1_missing_ef_money$code)# Look at the evolution over the years for the countries that have missing valuesEvol_Missing_ef_money <-ggplot(data = question1_missing_ef_money) +geom_point(aes(x = year, y = ef_money, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of economiv freedom: money over time", x ="Years from 2000 to 2022", y ="ef_money") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4) +scale_x_continuous(breaks =NULL)print(Evol_Missing_ef_money)# Linear interpolation for "ARM", "BFA", "BIH", "GEO", "KAZ", "LSO", "MDA", "MKD"list_code <-c("ARM", "BFA", "BIH", "GEO", "KAZ", "LSO", "MDA", "MKD")for (i in list_code) {# Filter the dataset for the current country country_data <- data_question1 %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$ef_money)# Update the original dataset with the interpolated values data_question1[data_question1$code == i, "ef_money"] <- interpolated_data}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values using the median by region.
# Create a dataframe that only have the countries with missing values and add a column which contains the % of missings for each countryquestion1_missing_ef_money <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_money))) %>%ungroup() %>%group_by(region) %>%filter(sum(PercentageMissing, na.rm =TRUE) >0)# See the distribution of the missings per regionFreq_Missing_ef_money <-ggplot(data = question1_missing_ef_money) +geom_histogram(aes(x = ef_money, fill =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%"))),bins =30) +labs(title ="Distribution of economic freedom: money", x ="ef_money", y ="Frequency") +scale_fill_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red","30-100%"="black"), labels =c("0-10%", "10-20%", "20-30%","30-100%")) +guides(fill =guide_legend(title ="% missings")) +facet_wrap(~ region, nrow =3)print(Freq_Missing_ef_money)# All are skewed distributions, we decide to replace the missing values where there are less than 30% missing by the median by regiondata_question1 <- data_question1 %>%group_by(code) %>%mutate(PercentageMissingByCode =mean(is.na(ef_money)) ) %>%ungroup() %>%# Remove grouping temporarilygroup_by(region) %>%mutate(MedianByRegion =median(ef_money, na.rm =TRUE),ef_money =ifelse( PercentageMissingByCode <0.3&!is.na(ef_money), ef_money,ifelse(PercentageMissingByCode <0.3, MedianByRegion, ef_money) ) ) %>%select(-PercentageMissingByCode, -MedianByRegion)
Economic freedom: trade
19 countries have missing values, but the percentage of missing values is always below 25%.
We look at the evolution of the variable over time. For the countries where this evolution is linear, we fill in the missing values using linear interpolation.
# Look at the evolution over the years for the countries that have missing valuesEvol_Missing_ef_trade <-ggplot(data = question1_missing_ef_trade) +geom_point(aes(x = year, y = ef_trade, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of economic freedom: trade over time", x ="Years from 2000 to 2022", y ="ef_trade") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4) +scale_x_continuous(breaks =NULL)print(Evol_Missing_ef_trade)# Linear interpolation for "AZE", "BFA", "ETH", "GEO", "VNH"list_code <-c("AZE", "BFA", "ETH", "GEO", "VNH")for (i in list_code) {# Filter the dataset for the current country country_data <- data_question1 %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$ef_trade)# Update the original dataset with the interpolated values data_question1[data_question1$code == i, "ef_trade"] <- interpolated_data}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values using the median by region.
# Create a dataframe that only have the countries with missing values and add a column which contains the % of missings for each countryquestion1_missing_ef_trade <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_trade))) %>%ungroup() %>%group_by(region) %>%filter(sum(PercentageMissing, na.rm =TRUE) >0)# See the distribution of the missings per regionFreq_Missing_ef_trade <-ggplot(data = question1_missing_ef_trade) +geom_histogram(aes(x = ef_trade, fill =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%"))),bins =30) +labs(title ="Distribution of economic freedom: trade", x ="ef_trade", y ="Frequency") +scale_fill_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red","30-100%"="black"), labels =c("0-10%", "10-20%", "20-30%","30-100%")) +guides(fill =guide_legend(title ="% missings")) +facet_wrap(~ region, nrow =3)print(Freq_Missing_ef_trade)# All are skewed distributions, we decide to replace the missing values where there are less than 30% missing by the median by regiondata_question1 <- data_question1 %>%group_by(code) %>%mutate(PercentageMissingByCode =mean(is.na(ef_trade)) ) %>%ungroup() %>%# Remove grouping temporarilygroup_by(region) %>%mutate(MedianByRegion =median(ef_trade, na.rm =TRUE),ef_trade =ifelse( PercentageMissingByCode <0.3&!is.na(ef_trade), ef_trade,ifelse(PercentageMissingByCode <0.3, MedianByRegion, ef_trade) ) ) %>%select(-PercentageMissingByCode, -MedianByRegion)
Economic freedom: regulation
12 countries have missing values, but the percentage of missing values is always below 25%.
We look at the evolution of the variable over time. For the countries where this evolution is linear, we fill in the missing values using linear interpolation.
# Create a dataframe that only have the countries with missing values and # add a column which contains the % of missings for each countryquestion1_missing_ef_regulation <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_regulation))) %>%filter(code %in% question1_missing_ef_regulation$code)# Look at the evolution over the years for the countries that have missing valuesEvol_Missing_ef_regulation <-ggplot(data = question1_missing_ef_regulation) +geom_point(aes(x = year, y = ef_regulation, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of economic freedom: regulation over time", x ="Years from 2000 to 2022", y ="ef_regulation") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4)print(Evol_Missing_ef_regulation)# Linear interpolation for "ETH", "KAZ", "MDA", "SRB"list_code <-c("ETH", "KAZ", "MDA", "SRB")for (i in list_code) {# Filter the dataset for the current country country_data <- data_question1 %>%filter(code == i)# Perform linear interpolation for the current country's data interpolated_data <-na.interp(country_data$ef_regulation)# Update the original dataset with the interpolated values data_question1[data_question1$code == i, "ef_regulation"] <- interpolated_data}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values using the median by region.
# Create a dataframe that only have the countries with missing values and # add a column which contains the % of missings for each countryquestion1_missing_ef_regulation <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_regulation))) %>%ungroup() %>%group_by(region) %>%filter(sum(PercentageMissing, na.rm =TRUE) >0)# See the distribution of the missings per regionFreq_Missing_ef_regulation <-ggplot(data = question1_missing_ef_regulation) +geom_histogram(aes(x = ef_regulation, fill =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%"))),bins =100) +labs(title ="Distribution of economic freedom: regulation", x ="ef_regulation", y ="Frequency") +scale_fill_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red","30-100%"="black"), labels =c("0-10%", "10-20%", "20-30%","30-100%")) +guides(fill =guide_legend(title ="% missings")) +facet_wrap(~ region, nrow =3)print(Freq_Missing_ef_regulation)# All are skewed distributions, we decide to replace the missing values where there are less than 30% missing by the median by regiondata_question1 <- data_question1 %>%group_by(code) %>%mutate(PercentageMissingByCode =mean(is.na(ef_regulation)) ) %>%ungroup() %>%# Remove grouping temporarilygroup_by(region) %>%mutate(MedianByRegion =median(ef_regulation, na.rm =TRUE),ef_regulation =ifelse( PercentageMissingByCode <0.3&!is.na(ef_regulation), ef_regulation,ifelse(PercentageMissingByCode <0.3, MedianByRegion, ef_regulation) ) ) %>%select(-PercentageMissingByCode, -MedianByRegion) %>%ungroup()na_count <-sapply(data_question1, function(x) sum(is.na(x)))print(na_count)#> X code #> 0 0 #> year country #> 0 0 #> continent region #> 0 0 #> population overallscore #> 0 0 #> goal1 goal2 #> 105 0 #> goal3 goal4 #> 0 0 #> goal5 goal6 #> 0 0 #> goal7 goal8 #> 0 0 #> goal9 goal10 #> 0 105 #> goal11 goal12 #> 0 0 #> goal13 goal15 #> 0 0 #> goal16 goal17 #> 0 0 #> unemployment.rate GDPpercapita #> 0 0 #> MilitaryExpenditurePercentGDP internet_usage #> 0 0 #> pf_law pf_security #> 0 0 #> pf_movement pf_religion #> 0 0 #> pf_assembly pf_expression #> 0 0 #> pf_identity ef_government #> 0 0 #> ef_legal ef_money #> 0 0 #> ef_trade ef_regulation #> 0 0
Now, we notice that there were only missing values for goals 1 and 10. As we did before, we have started to investigate where are located the NAs in our dataset for first goal1, then goal 10.
# goal1question1_missing_goal1 <- data_question1 %>%group_by(code) %>%summarize(Na_goal1 =mean(is.na(goal1)))%>%filter(Na_goal1 !=0)data_question1 <- data_question1 %>%filter(!code %in% question1_missing_goal1$code)# Update List of countries deletedlist_country_deleted <-c(list_country_deleted, "KWT","NZL","OMN","SGP","UKR")#still 42 NA values goal10
We had found that the missing values were located in only 5 countries. So we have decided to get rid of them. At this stage, there were only 42 remaining missing values. Then, same step for goal 10.
#goal10question1_missing_goal10 <- data_question1 %>%group_by(code) %>%summarize(Na_goal10 =mean(is.na(goal10)))%>%filter(Na_goal10 !=0)data_question1 <- data_question1 %>%filter(!code %in% question1_missing_goal10$code)# Update List of countries deletedlist_country_deleted <-c(list_country_deleted, "GUY","TTO")
We have found the 2 lasts contries containing missing values. Now, our dataset is completely clean and ready to be used for our question 1.
Data for question 2 and 4
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. Since there are no other missing values, we stop here.
#### Questions 2 and 4see_missing24 <- data_question24 %>%group_by(code) %>%summarise(across(everything(), ~sum(is.na(.))) %>%mutate(num_missing =rowSums(across(everything()))) %>%filter(num_missing >0))# Nothing to remove, only goals 1 and 10 have missing (already discussed before)
Data for question 3
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. Since there are no other missing values, we stop here.
Disasters
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. We find out that there are many missing values
In this particular case, even it there are many missing values in our disaster dataset, we made the hypothesis that disaster events can not happen every year for every country given that these are uncontrollable and non-recurring events. Therefore the NAs that we encounter will become zeroes, implying that there have been no climatic disasters.
COVID19
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. Since there are no other missing values, we stop here.
# COVIDsee_missing3_2 <- data_question3_2 %>%group_by(code) %>%summarise(across(-c(goal1, goal10), # Exclude columns "goal1" and "goal10"~sum(is.na(.))) %>%mutate(num_missing =rowSums(across(everything()))) %>%filter(num_missing >0))# No missing
Conflicts
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed.Two countries have missing values, we remove them (MNE and SRB).
# Conflictssee_missing3_3 <- data_question3_3 %>%group_by(code) %>%summarise(across(-c(goal1, goal10), # Exclude columns "goal1" and "goal10"~sum(is.na(.))) %>%mutate(num_missing =rowSums(across(everything()))) %>%filter(num_missing >0))# 2 countries have missings, we remove them: MNE and SRBdata_question3_3 <- data_question3_3 %>%filter(!code %in%c("MNE","SRB"))##### EXPORT as CSV #####write.csv(data_question1, file =here("scripts","data","data_question1.csv"))write.csv(data_question24, file =here("scripts","data","data_question24.csv"))write.csv(data_question3_1, file =here("scripts","data","data_question3_1.csv"))write.csv(data_question3_2, file =here("scripts","data","data_question3_2.csv"))write.csv(data_question3_3, file =here("scripts","data","data_question3_3.csv"))